Luminous Landscape Forum
Raw & Post Processing, Printing => Colour Management => Topic started by: Jack Hogan on March 19, 2018, 04:20:47 am
-
Hello Color Gurus, I have a question.
Say you are an astrophotographer at a dark site looking at a patch of sky through a telescope inside a hut in complete darkness. There is no light pollution so light from the telescope's narrow field of view strikes your eye and you perceive the wondrous colors of galaxies and nebulae within it. With some effort and the proper tools you are able to estimate the spectral power distribution of a particular nebula. You would like to calculate what coordinates the relative tone corresponds to in an output color space it fits in - let's say Adobe RGB - to check against the same captured and rendered by a digital camera. This is what I have been doing to obtain Adobe RGB values calculated from spectrum- but I have a question in the back of my head about Observer Adaptation that's been nagging at me for a while:
1) dot product of absolute spectral irradiance of nebula by the CIE CMF of choice;
2) Adaptation to D65
3) projection to Adobe RGB
4) application of gamma.
What is the starting white point for calculating the adaptation in 2)? Is it illuminant E, because CIE CMFs integrate to 1; or is it the background mix of galaxies and stars in the telescope's field of view that I assume the observer is adapted to?
Thanks for any pointers,
Jack
-
I doubt standard CIE CMF will be useful in such case.
-
Hi Marcin,
For the sake of argument, then, pretend that the absolute spectral irradiance from the nebula is the same as that reflected from the blue patch in a ColorChecker 24 illuminated by a star (the sun). Say that same absolute spectral irradiance is in the field of view of the telescope in the situation I described. How would you estimate its Adobe RGB coordinates then?
Jack
-
Interesting because it might depend on the incidence of the light received from the object. That is, one would probably like to know the color relative to the sun, thus relative to the sun-atmosphere interaction, meaning the reference should be a normal daylight equivalent. However, the corresponding temperature likely depends on where the object is located relative to the horizon.
-
There is no light pollution so light from the telescope's narrow field of view strikes your eye and you perceive the wondrous colors of galaxies and nebulae within it. With some effort and the proper tools you are able to estimate the spectral power distribution of a particular nebula.
Note that color perception depends on the light level. Natural night time viewing will be in the scotopic or mesopic range, so normal observer spectral sensitivities don't apply. In the scotopic range, vision is monochromatic.
So a key question is whether you are trying to reproduce color as we would naturally see it, or whether you are attempting something more synthetic - i.e. how would we see it if the light levels were in the photopic range ?
(You may find this website (http://www.clarkvision.com/articles/nightscapes/) of interest. )
-
Interesting because it might depend on the incidence of the light received from the object. That is, one would probably like to know the color relative to the sun, thus relative to the sun-atmosphere interaction, meaning the reference should be a normal daylight equivalent. However, the corresponding temperature likely depends on where the object is located relative to the horizon.
Hello Oscar, thanks for your thoughts.
In this case the sun per se would not be in the field of view but perhaps a number of sun-like stars and other faint background lighting might be (so called sky glow, what the Hubble would see).
I am thinking that if we knew the spectrum of three or more objects (e.g. nebulae or galaxies) in the field of view with 'orthogonal' colors and we captured them with a digital camera we could easily calculate a compromise color matrix from raw data to linear Adobe RGB using CIE CMFs to provide a reference. CIE CMFs produce absolute values in XYZ with white point 'E', so I guess they assume that the observer is adapted to illuminant E - and that should be the starting adaptation to rotate to D65 for Adobe RGB. Does this make sense?
-
(You may find this website of interest. )
Hi Graeme,
Funny you should mention that: the color pages on that site are the reason I got thinking about the question in the OP in the first place. They are wildly (did I say wildly? WILDLY!) wrong.
To calculate colors in Adobe RGB from spectrum he dot multiplies it with Stiles and Burch 1955 RGB CMFs (the ones with negative portions) and stores the result as-is in an Adobe RGB file. No conversion to XYZ, no adaptation to D65, no projection to Adobe RGB, no gamma (!). For instance, take a look at Figures 5-8 here (http://www.clarkvision.com/articles/astrophotography.m45-pleiades.true.color/), not a single one of those tones is displayed properly (colorimetrically).
It became apparent very quickly in this thread (https://www.dpreview.com/forums/post/60879079) that Roger Clark does not know what he is doing as far as displaying 'true' color is concerned. His relative pages are an embarrassment and should be corrected or taken down. I offered to help him do them right but so far he has not taken me up on it (:-).
Note that color perception depends on the light level. Natural night time viewing will be in the scotopic or mesopic range, so normal observer spectral sensitivities don't apply. In the scotopic range, vision is monochromatic. So a key question is whether you are trying to reproduce color as we would naturally see it, or whether you are attempting something more synthetic - i.e. how would we see it if the light levels were in the photopic range ?
Yes, that's something else to think about. Since human observers actually do see colors of celestial objects through powerful telescopes I think we are at least in mesopic vision, if not full fledged photopic. But for the sake of this question let's assume that the telescope is capable of collecting enough light (> a few cd/m^2) that the Observer is in the Photopic range where CIE CMFs apply.
So, when converting from spectrum to Adobe RGB, can I assume a starting adaptation of illuminant 'E' once CIE CMFs have landed me in XYZ? Or do I have to take Observer adaptation to the background in the telescope's field of view into consideration? After all, the original CMFs were derived by projecting absolute spectral irradiances (of which the nebula in question is an example), no?
Jack
-
It became apparent very quickly in this thread (https://www.dpreview.com/forums/post/60879079) that Roger Clark does not know what he is doing as far as displaying 'true' color is concerned. His relative pages are an embarrassment and should be corrected or taken down. I offered to help him do them right but so far he has not taken me up on it (:-).
Ah - sad to hear, since the website claims to be "doing it right"!
So, when converting from spectrum to Adobe RGB, can I assume a starting adaptation of illuminant 'E' once CIE CMFs have landed me in XYZ?
Probably not, although it can be hard to figure out what the observer will take as white with unconventional scenes, and some conventional white points like D50 and E would be worth experimenting with in comparison with other approaches.
Or do I have to take Observer adaptation to the background in the telescope's field of view into consideration?
Worth a try.
It might also be worth following up some of Thor Olson's articles in the CIC proceedings:
High Dynamic Range Astronomical Imaging (http://www.nightscapes.net/Notes/HDR/HDR for Astronomical Imaging.CIC-15.pdf)
The Colors of the Stars (http://ww.nightscapes.net/techniques/TechnicalPapers/StarColors.pdf)
The Colors of the Deep Sky (https://www.researchgate.net/profile/Thor_Olson/publication/221501779_The_Colors_of_the_Deep_Sky/links/09e415080d658b2a99000000/The-Colors-of-the-Deep-Sky.pdf)
-
I don't think adaptation in the usual sense is applicable in this case. I don't believe our perception adapts to background radation visible through the telescope, or at least we can ignore that. If we can also ignore atmospherics then perhaps it's more a question of what to consider white, and particularly what colortemperature. Then you could use temperature to blackbody spectrum conversion as the source illuminant.
Hence, you could take the sun as a reference for a white star (even though it is obviously not in the frame), and "adapt" all spectral data of other stars and objects using the sun's corresponding blackbody spectrum (T = ~5800 something) and convert to XYZ.
The more pressing problem likely is how to "ignore" the background radiation accumulated during an exposure. I remember having written a plugin to compensate for inverse fall-off, and vaguely recall it wasn't a simple subtraction. I'm sure though there must be some really useful and open solutions for that problem these days which may also help explain the best options for the illuminant problem.
Perhaps it turns out to be more of an easthatic problem than it is a scientific one? i.e. the nightsky may not actually be blue and the stars are perhaps mostly white, but we just prefer the sky darkblue and the stars yellowish...
-
I don't think adaptation in the usual sense is applicable in this case. I don't believe our perception adapts to background radation visible through the telescope, or at least we can ignore that. If we can also ignore atmospherics then perhaps it's more a question of what to consider white, and particularly what colortemperature. Then you could use temperature to blackbody spectrum conversion as the source illuminant.
Hence, you could take the sun as a reference for a white star (even though it is obviously not in the frame), and "adapt" all spectral data of other stars and objects using the sun's corresponding blackbody spectrum (T = ~5800 something) and convert to XYZ.
The more pressing problem likely is how to "ignore" the background radiation accumulated during an exposure. I remember having written a plugin to compensate for inverse fall-off, and vaguely recall it wasn't a simple subtraction. I'm sure though there must be some really useful and open solutions for that problem these days which may also help explain the best options for the illuminant problem.
Perhaps it turns out to be more of an easthatic problem than it is a scientific one? i.e. the nightsky may not actually be blue and the stars are perhaps mostly white, but we just prefer the sky darkblue and the stars yellowish...
If adaptation in the usual sense is not applicable in such case, what can explain Kruithof curve/Purkinje effect? The stars are monochromatic (or "white") when their brightness is in scotopic range, the sky is darkblue when its brightness is in mesopic range.
-
If adaptation in the usual sense is not applicable in such case, what can explain Kruithof curve/Purkinje effect? The stars are monochromatic (or "white") when their brightness is in scotopic range, the sky is darkblue when its brightness is in mesopic range.
But that concerns adaptation to darkness, not adaptation to colordifferences of illuminants. I think the background in a telescope or perhaps even in a long exposure is simply not judged by our perception as a dominant illuminant color, even if brighter objects are available to trigger photopic vision.
-
Ah - sad to hear, since the website claims to be "doing it right"!Probably not, although it can be hard to figure out what the observer will take as white with unconventional scenes, and some conventional white points like D50 and E would be worth experimenting with in comparison with other approaches.Worth a try.
It might also be worth following up some of Thor Olson's articles in the CIC proceedings:
High Dynamic Range Astronomical Imaging (http://www.nightscapes.net/Notes/HDR/HDR for Astronomical Imaging.CIC-15.pdf)
The Colors of the Stars (http://ww.nightscapes.net/techniques/TechnicalPapers/StarColors.pdf)
The Colors of the Deep Sky (https://www.researchgate.net/profile/Thor_Olson/publication/221501779_The_Colors_of_the_Deep_Sky/links/09e415080d658b2a99000000/The-Colors-of-the-Deep-Sky.pdf)
Thanks Graeme, very interesting articles, especially the last one. Unfortunately he does not show his work in obtaining the final matrices, so I am not sure what his starting (white) point ended up being.
-
I don't think adaptation in the usual sense is applicable in this case. I don't believe our perception adapts to background radation visible through the telescope, or at least we can ignore that. If we can also ignore atmospherics then perhaps it's more a question of what to consider white, and particularly what colortemperature. Then you could use temperature to blackbody spectrum conversion as the source illuminant.
Hence, you could take the sun as a reference for a white star (even though it is obviously not in the frame), and "adapt" all spectral data of other stars and objects using the sun's corresponding blackbody spectrum (T = ~5800 something) and convert to XYZ.
The more pressing problem likely is how to "ignore" the background radiation accumulated during an exposure. I remember having written a plugin to compensate for inverse fall-off, and vaguely recall it wasn't a simple subtraction. I'm sure though there must be some really useful and open solutions for that problem these days which may also help explain the best options for the illuminant problem.
Perhaps it turns out to be more of an easthatic problem than it is a scientific one? i.e. the nightsky may not actually be blue and the stars are perhaps mostly white, but we just prefer the sky darkblue and the stars yellowish...
All good points Oscar.
-
Allow me to approach the problem from a different direction:
Say I have been sitting in a dark room for a while, totally dark except for the power LED on the TV (say red, narrowband, around 700nm but within Adobe RGB's gamut). I am dark adapted but I see the normally bright LED well and recognize its color as the 'correct' shade of red (correct in the sense that it looks to me to be about the same color as when the lights are on).
A) Now say I want to calculate the Adobe RGB coordinates of the LED directly from its spectrum. I take the 700nm narrowband SPD and dot-multiply it with the CIE CMFs of choice. Do I need to adapt to D65? If so, from what white point?
B) Next I capture the dark scene with the LED in it with a DSLR. The Adobe RGB value a linear rendering will produce will be within 1 dE2000 as long as the choice of WB/matrix is consistent. I simulated this for white points in the 4000-7000K range, each time recalculating the WB and Compromise Color Matrix as I normally do - minimizing a color error function to reference LAB values of a CC24, obtained via CIE CMFs. So the camera is effectively being used as a colorimeter and it produces relatively constant, illuminant-independent, absolute values in Adobe RGB for the LED. Will the values in A) and B) match? And am I leading the witness, by finding the CCM as I do?
Jack
-
But that concerns adaptation to darkness, not adaptation to colordifferences of illuminants. I think the background in a telescope or perhaps even in a long exposure is simply not judged by our perception as a dominant illuminant color, even if brighter objects are available to trigger photopic vision.
Exactly, but the question is what's the state of chromatic adaptation when we're adapted to darkness? IMO the Purkinje effect may indicate, that it's not at D50-E range in case of such viewing condition. Furthermore, darker objects in the image will be desaturated or monochromatic, so when we use standard CMF the resulting image may be oversaturated in nearblacks and shadows.
-
Exactly, but the question is what's the state of chromatic adaptation when we're adapted to darkness? IMO the Purkinje effect may indicate, that it's not at D50-E range in case of such viewing condition. Furthermore, darker objects in the image will be desaturated or monochromatic, so when we use standard CMF the resulting image may be oversaturated in nearblacks and shadows.
I agree Marcin, but let's assume we are not interested in what else is in the room. We just want to know what the coordinates should be in Adobe RGB to reproduce the LED's color 'accurately'. Perhaps we should assume that the monitor displaying the captured LED is in a pitch dark room? How do we go from absolute spectral irradiance to absolute XYZ values to the Adobe RGB file in that case? If the camera can do it, why can't we calculate it empirically?
-
Exactly, but the question is what's the state of chromatic adaptation when we're adapted to darkness? IMO the Purkinje effect may indicate, that it's not at D50-E range in case of such viewing condition. Furthermore, darker objects in the image will be desaturated or monochromatic, so when we use standard CMF the resulting image may be oversaturated in nearblacks and shadows.
Perhaps under non-photopic vision our colorperception can be considered a system at rest in default state; i.e. no excitation = no adaptation so it might be at rest representing the overall average which probably is some kind of daylight equivalent. Saturation, especially in near-blacks, will certainly be interesting, first because we probably prefer saturated objects in this context, and second because we don't actually know what saturation would be deemed "normal" since we don't "normally" view the saturated version as you mentioned (other than artist's impressions of course)...
-
Allow me to approach the problem from a different direction:
Say I have been sitting in a dark room for a while, totally dark except for the power LED on the TV (say red, narrowband, around 700nm but within Adobe RGB's gamut). I am dark adapted but I see the normally bright LED well and recognize its color as the 'correct' shade of red (correct in the sense that it looks to me to be about the same color as when the lights are on).
A) Now say I want to calculate the Adobe RGB coordinates of the LED directly from its spectrum. I take the 700nm narrowband SPD and dot-multiply it with the CIE CMFs of choice. Do I need to adapt to D65? If so, from what white point?
B) Next I capture the dark scene with the LED in it with a DSLR. The Adobe RGB value a linear rendering will produce will be within 1 dE2000 as long as the choice of WB/matrix is consistent. I simulated this for white points in the 4000-7000K range, each time recalculating the WB and Compromise Color Matrix as I normally do - minimizing a color error function to reference LAB values of a CC24, obtained via CIE CMFs. So the camera is effectively being used as a colorimeter and it produces relatively constant, illuminant-independent, absolute values in Adobe RGB for the LED. Will the values in A) and B) match? And am I leading the witness, by finding the CCM as I do?
Jack
I think in this case you can safely assume a daylight equivalent for the source, and it might as well be D65 to make it simple.
The horror of actually displaying the correct representation is however in the colormanagement details of your viewing system. If you don't properly prepare the result as source for the displaysystem, you still end up with skewed results.
What if it was a "white" LED? How would you want to represent that particular white LED in your destination space and on your monitor? Should it be R=G=B=1 in both spaces?
-
I think in this case you can safely assume a daylight equivalent for the source, and it might as well be D65 to make it simple.
Thanks Oscar - good point about us having a 'resting' adaptation around daylight CCTs. Do you see any counter indications to using 'E'? That's what I have used in the past and at 5450K it seems to be right in the thick of things.
The horror of actually displaying the correct representation is however in the colormanagement details of your viewing system. If you don't properly prepare the result as source for the displaysystem, you still end up with skewed results.
What if it was a "white" LED? How would you want to represent that particular white LED in your destination space and on your monitor? Should it be R=G=B=1 in both spaces?
Tricky question. What white? Say 'E'. It would define the white point, would it not? Then it would be just a matter of rotating XYZ coordinates obtained the usual way to D65 and projecting to Adobe RGB (where R=G=B=1 for white).
-
Thanks Oscar - good point about us having a 'resting' adaptation around daylight CCTs. Do you see any counter indications to using 'E'? That's what I have used in the past and at 5450K it seems to be right in the thick of things.
I personally like E very much, especially for emissive cases as discussed here. For reflective cases it might be debatable.
I also wouldn't be surprised if the chromaticity coordinates of E are the resting state of adaptation, since it can be considered an optimal spot to adapt from. Mind you, I have no proof or research whatsoever to substantiate any of this.
-
Interesting questions, Jack. In my experience, what I see -- or saw, since it's been a long time -- when looking through a 14 inch or smaller telescope (I've never had the pleasure of riding high on a 200 inch one) was dramatically different from what the professional astronomers turned out. Even with a 14-incher, there are a lot of objects that you look at out of the corner of your eye to make them appear bright enough, and now you're using strictly scotopic vision (which is why you perform that little trick in the first place). Even bright nebulae look pale through a small scope compared to the shots the big boys (or the Hubble) take. I've also noticed that when looking at the Milky Way with bare eyes that it looks a lot paler than all the shots that I see now that digital photography has made taking images of our galaxy falling-off-a-log easy.
So I question the whole idea of the viewer's adaptation when looking through a telescope as the basis for adjusting the colors in astro images. I think they are what they are in terms of color, and any way that you get them to look "right" (and right is now defined for many as "like all the other astro images, and maybe in particular by the ones from the Hubble) is fair game.
Jim
-
Good comments Jim, even though they tend to defeat the purpose of this thread :)
I was surprised to learn that most astrophotographers process their images for maximum impact, they don't even try to show approximately accurate tones (including many images out of the Hubble). For instance, they stretch contrast logarithmically and often use faux color for emphasis so their objective has more to do with emphasizing different features than anything else. Someone read one of the posts that documents what I am learning about color and asked me to comment on someone else's page that claimed to provide 'true' astro colors that looked wrong to this person. It was clear right away that that page was based on an incorrect understanding of CMFs thus resulting in grossly incorrect tones no matter the underlying assumptions. However it got me thinking about what it would take to do it right, hence the question on adaptation in this thread.
The concept of 'resting' adaptation, to use Oscar's terminology, is an interesting one. Here is the question again, slightly reformulated:
Say I have been sitting in a dark room, totally dark except for the power LED on the TV (say red, narrowband, around 600nm, just inside Adobe RGB's gamut). I am dark adapted but I see the normally bright LED well and recognize its color as the 'correct' shade of red (correct in the sense that it looks to me to be roughly the same red as when there is light in the room).
Now say I want to calculate the Adobe RGB coordinates of the LED directly from its spectrum. I take the 600nm narrowband SPD and dot-multiply it by the 1931 2 deg CIE XYZ CMFs of choice*. Do I need to adapt to D65 before projecting to Adobe RGB? If so, from what white point?
In other words what's our 'resting' adaptation when there is no ambient light?
What do you think?
Jack
*My impression by looking at said LED is that it looks somewhat more saturated in the dark than when there is light in the room but the hue is about right - so it would seem that CIE's photopic CMFs is possibly not too far off in this situation.
-
The concept of 'resting' adaptation, to use Oscar's terminology, is an interesting one.
I've been assure by someone who should know (Mark Fairchild) that given long enough, you will adapt to any illuminant.
Given that the underlying chromatic adaptation mechanism is assumed to be the independent light level adaptation of the L, M & S cones, this makes sense. It would also explain why most individuals have a similar sense of what "white" is, even though there is significant individual variation in the density of L, M & S cones in the retina.
So the expected answer if you wait long enough, is that there is no such thing as a 'resting' adapation.
-
I've been assure by someone who should know (Mark Fairchild) that given long enough, you will adapt to any illuminant.
Yeah, but the problem at hand is "no illuminant".
Given that the underlying chromatic adaptation mechanism is assumed to be the independent light level adaptation of the L, M & S cones, this makes sense. It would also explain why most individuals have a similar sense of what "white" is, even though there is significant individual variation in the density of L, M & S cones in the retina.
Well, we've recently seen some internet images with a dress and a shoe that suggest this doesn't entirely hold.
In addition, what I find most intriguing in relation to this is the workings of auto-white-balance and metering on digital camera's. It's very easy to test that the average color of the entire image is not a good indicator of perceived white-balance. Neither linear space average, nor perceptual average. Similarly for median. There appears to be something happening in our perception that prioritises bright areas or perhaps bright peaks for whitebalance determination and thus possibly also for adaptation.
-
I've been assure by someone who should know (Mark Fairchild) that given long enough, you will adapt to any illuminant.
I haven't seen Mark in a long time -- say "Hi" to him for me, please -- but I can tell you that after two hours straight in the darkroom under a Thomas sodium vapor safelight, it still looks yellow. Maybe to get adapted to that, I'd have to bring in food and a cot.
Jim
-
.
*My impression by looking at said LED is that it looks somewhat more saturated in the dark than when there is light in the room but the hue is about right - so it would seem that CIE's photopic CMFs is possibly not too far off in this situation.
That experiment combines changing the luminance of the surround as well as its chromaticity, doesn't it?
Jim
-
*My impression by looking at said LED is that it looks somewhat more saturated in the dark than when there is light in the room but the hue is about right - so it would seem that CIE's photopic CMFs is possibly not too far off in this situation.
That experiment combines changing the luminance of the surround as well as its chromaticity, doesn't it?
Yes. I was thinking that in addition to adaptation two other effects might be at play: contrast and desaturation. We know even from post processing that increased contrast typically makes things look more colorful (think fireworks at night). And we also know that adding more 'white' illuminant light to a patch of color desaturates it (think chromaticity diagram).
Jack
-
I haven't seen Mark in a long time -- say "Hi" to him for me, please -- but I can tell you that after two hours straight in the darkroom under a Thomas Sodium vapor safelight, it still looks yellow. Maybe to get adapted to that, I'd have to bring in food and a cot.
Jim
Could be a cognitive effect. See pg. 149 of Mark Fairchild's book "Color Appearance Models"
-
-- but I can tell you that after two hours straight in the darkroom under a Thomas sodium vapor safelight, it still looks yellow. Maybe to get adapted to that, I'd have to bring in food and a cot.
Mark implied that it may well take a very long time i.e. possibly days :-)
I guess it may involve a level of neurological adaptation for extreme illuminant hues, analogous to the "mirror glasses" type experiments (https://www.theguardian.com/education/2012/nov/12/improbable-research-seeing-upside-down) where the subjects report seeing the world upside down for some time, until it flips.
(I chatted to him about this specific subject at a recent CIC, as it has a bearing on something I've been working on.)